Sarcasm Detection in News Headlines

This project focuses on detecting sarcasm in news headlines by utilizing a combination of machine learning and deep learning models. The aim is to enhance the accuracy of sentiment analysis systems by addressing the challenge posed by sarcasm. The dataset used is sourced from Kaggle, which contains labeled news headlines. The proposed solution includes training two separate models: a stacking model (using SVM, Random Forest, and XGBoost) combined with LSTM, and a BERT-based model. The stacking model with LSTM captures both structural and contextual features of sarcasm, while the BERT model leverages pre-trained embeddings for better text representation. The system is deployed through a Flask backend with a user-friendly interface. The project evaluates both models for sarcasm detection and compares their performances.

Keywords: Sarcasm detection, news headlines, stacking, LSTM, BERT, sentiment analysis, machine learning, deep learning, sarcasm classification, Flask.